Postgraduate research opportunities

The value of automation for infrastructure decision making

This PhD project will quantify the value of automated monitoring when managing assets. The framework developed will help asset owners decide whether new techniques in robotic sensor deployment and machine learning.

Number of places

1

Funding

International fee, Home fee, Equipment costs, Stipend

Opens

27 January 2020

Deadline

27 March 2020

Duration

42 months

Eligibility

Applicants should have or expect a distinction pass at Master’s level, or a first class/ 2:1 BEng/BSc Honours degree in an Engineering, Physical Sciences or Management Sciences subject. Candidates with strong analytical and interpersonal skills are particularly welcome to apply. Experience with risk analysis, structural computational modelling, or programming is advantageous but not essential.

Eligibility for RCUK studentships

  • Research Council (RC) fees and stipend can only be awarded to UK and EU students and not to EEA or International students.
  • EU students are only eligible for RC stipend if they have been resident in the UK for 3 years, including for study purposes, immediately prior to starting their PhD.
  • If an EU student cannot fulfil this condition then they are eligible for a fees only studentship.
  • International students cannot be funded from RC funds unless they are ‘settled’ in the UK. ‘Settled’ means being ordinarily resident in the UK without any immigration restrictions on the length of stay in the UK. To be ‘settled’ a student must either have the Right to Abode or Indefinite leave to remain in the UK or have the right of permanent residence in the UK under EC law. If the student’s passport describes them as a British citizen they have the Right of Abode.
  • Students with full Refugee status are eligible for fees and stipend.

Project Details

Motivation: Across engineering sectors, current practice for asset integrity assessment typically relies on human inspection, coupled with a few manually fitted sensors. The widespread implementation of practical / data automation faces reluctance, partly due to costs (which will change), but partly due to a “decision problem” — a need for transparent, defensible decision making (this will not change).

Automated monitoring systems, installed with robotics and/or interpreted with black box machine learning techniques, will result in different data (shown as a blue circle) than conventional manual inspections (purple). Regardless of which path is chosen, the purpose of the data is to answer a question: do we take action (e.g. maintenance) or do we wait? In reality, this is a multi-criteria assessment of the trade-offs between costs, safety and the defensibility of the decision making process. For manual inspection, this decision making process is familiar and transparent, but it is also expensive and prone to human errors that are difficult to trace.

Outcomes: The main outcome of this project will be a framework that helps asset owners decide on their level of commitment to automated monitoring. The inputs into the framework will be:

1. models / knowledge that describe the state / degradation of the asset over time;

2. an assessment of the relative efficacy of automated and manual inspection methods in diagnosing system state / integrity;

3. a process to elicit from key stakeholders the trade-offs between cost, safety and defensibility.

Item 1 is something that most asset managers already have or can develop. Items 2 and 3 will be delivered by this project.

Objectives: The project’s objectives are to:

  • instrument structural components in the lab with strain sensors, using both manual methods and robotics, and compare variability in sensor performance (placement location, strain transfer, precision, accuracy and reliability);
  • test the instrumented components under dynamic loading, and use the obtained data with an artificial neural network to update the parameters of a finite element (FE) model of the system;
  • use the updated FE to estimate the component’s fragility and infer resilience: where possible, connect this with a risk of downtime and a revenue loss.
  • using this framework, explore the knock-on effects that automation of sensor placement and data analysis have on the cost and value of information provided by monitoring
  • use expert interviews to explore how the defensibility of decision making is affected by automating the monitoring as proposed.

 

Funding Details

Fees, stipend and project expenses

Number of places

0

Further information

Sensors are relatively cheap, and so the major costs of a monitoring campaign are comprised of the staff time for planning, installation and analytics, disruption to service, and opportunity costs.

The value of the campaign, or the value of information (VoI) that it yields, is the amount of money that the asset owner would be willing to pay for the information prior to making their decisions. This is linked to variables like the quality of the data (e.g. its noise), trust, and how useful it is when optimising maintenance.

If the VoI exceeds the cost of getting it, then an asset manager will at least consider a monitoring campaign. Until now, the errors and costs associated with manually performing sensor installation and data analytics tasks have meant costs often win out. The automation of these tasks with robotics and machine learning are, however, disrupting this. It is not yet clear how. We know that our conventional understanding of cost, risk, data reliability and how easily we can understand where our results come from (explainability) are changing, but there are unanswered questions, e.g.:

  • How valuable is sensor data if acquired accurately by a robot rather than a person, and does this value exceed the investment costs?
  • How much less valuable or defensible is information when it is obtained using opaque, black-box techniques instead of “fully explainable” numerical models?

Thanks to finite element model updating using black-box techniques [1-3], fragility curves can be very quickly produced and updated based on sensor data. These fragility curves show how the damage state of an asset (slight, moderate, extensive, complete) affect its probability of failure during a hazard. If we have a higher confidence in the performance of robotically deployed sensors, for example, we can make less conservative judgements about asset failure. This translates into when we should spend money on maintenance actions: it essentially represents the value of robotic installation.

Once we have a recovery model (an idea of how quickly assets return to service after losing functionality [5]), we can convert fragility curves into a measure of asset resilience. Our group’s links with insurance companies may allow us to formalise this relationship between damage states and indirect losses due to downtime. Insurance claims related to losses due to business interruption are now recognized to be one of the most significant factors in loss generation. Traditional methods to assess direct losses are data based but data for indirect losses is sparse. Therefore, resilience considerations may provide the basis for assessing this. By quantifying the probability that the client will suffer resilience failure, the degree of insurance can be established [6].

The student would join the University of Strathclyde’s 60-credit postgraduate training programme leading to the Postgraduate Certificate in Researcher Professional Development.

 The student will benefit from interaction with other academics and PhD students within an active research community, as well as being embedded within the SMART Pumps for Subsurface Engineering Project, a joint EPSRC-industry funded multi-institutional partnership, and interacting and engaging with researchers at the School of Geosciences at the University of Edinburgh.

 References

[1] T Marwala, Finite-element-model Updating Using Computional Intelligence Techniques, available at: https://link.springer.com/content/pdf/10.1007%2F978-1-84996-323-7.pdf

[2] M Atalla, Model Updating Using Neural Networks, available at http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.3.846&rep=rep1&type=pdf

[3] S Cooper et al., Integration of system identification and finite element modelling of nonlinear vibrating structures, available at https://www.sciencedirect.com/science/article/pii/S0888327017305137

[4] I. Gidaris et al. “Multiple-hazard fragility and restoration models of highway bridges for regional risk and resilience assessment in the United States: State-of-the-art review”. In: Journal of Structural Engineering (United States) 143.3 (2017)

[5] A. Decò, P. Bocchini, and D.M. Frangopol. “A probabilistic approach for the prediction of seismic resilience of bridges”. In: Earthquake Engineering and Structural Dynamics 42.10 (2013), pp. 469–1487

[6] M. H. Faber , Insurance Risk and Resilience Modeling, JCSS PhD Course on: Structural Reliability and Probabilistic Model Code & Risk Informed Decision Making and Decision Analysis, 2019, Tongji, China 

Contact us

Dr Marcus Perry, m.perry@strath.ac.uk, 0141 548 4942

How to apply

Please contact Dr Marcus Perry, m.perry@strath.ac.uk, 0141 548 4942